Naming Scheme(s)

[Andy Bierman <abierman@cisco.com>]

Hi,

I want to fully understand the rationale for multiple naming schemes
for the data definition language. 

It seems the difference between SMI and SPPI naming is that SMI instance
names can have multiple components, of various (but not all) base types.
SPPI instance names consist of a single arbitrary local integer, and the
UNIQUENESS clause is used to identify the attributes that would normally
be index components in SMIv2.

I believe there should be a single naming scheme, because this is
really part of the data definition language, not the management protocol.
The term 'protocol-mapping' that the WG has been using is misleading.  
It is really an SMI or SPPI mapping that we are talking about. SNMP and COPS-PR
both claim to be independent of any data naming scheme, as long as the
OBJECT IDENTIFIER base type is used to name a data object.

Operational experience has shown that arbitrary local integers 
are not that useful to management applications.  Vendors end up
assigning proprietary algorithms to the namespace (e.g. RMON*-MIB 
control tables and ifIndex) so the integers aren't arbitrary anymore.
This leads to multi-manager problems when not every application
developer knows there's a hidden algorithm, and actually believes the
text in the MIB. In summary, it saves bytes in an OID at the expense
of interoperability. IMO, this practice should be deprecated, 
not encouraged.

SMIng should adopt a single naming scheme that is not based
on arbitrary integers. My first choice is SMI-DS naming, 
and my second choice is to keep SMIv2 naming.

Many details in the data definition language depend on this
decision, and it would be nice to figure it out before the IETF.

thanks,
Andy